In computational biomechanics, there are two well-developed but separate modeling domains: multi body dynamics for body movements, and finite element modeling for tissue deformations. Many clinical problems, however, span both domains. Whole body anatomy, mass distribution, and gait pattern are not typically represented in finite element models, yet these are important real-world factors that affect tissue stresses in the musculoskeletal system, which may contribute to clinical problems such as osteoarthritis and diabetic foot ulceration. Movement simulations, on the other hand, lack a representation of tissue deformations, which are indicators of mechanically induced pain and other sensory feedback (or the lack thereof) and will cause observable changes in gait. Exploration of these neuromusculoskeletal integrative mechanisms can only be accomplished by multi-domain simulations. Current techniques for multi-domain modeling are insufficient because forward dynamic movement simulations typically proceed along a sequence of many small steps in time. Finite element models are too slow to allow a solution at each of these steps. One may painstakingly produce a single movement simulation, but not the thousands of simulations that are required for predictive movement optimizations that are the state of the art in musculoskeletal dynamics. This has become a bottleneck for our own research, as well as for others. Our first aim, therefore, is to implement a generic, self-refining, surrogate modeling scheme, which aims to reproduce an underlying physics-based finite element model within a given error tolerance, but at a far lower computational cost. The self-refining feature is the key to reproduce the multi-dimensional input-output space of a typical finite element model of a joint or joint complex. Our second aim is to demonstrate the utility of these tools by connecting a finite element model of the foot to a complete musculoskeletal gait simulation, which will test the hypothesis that peak plantar pressures (an indicator of diabetic foot ulceration), can be lowered under safety thresholds by selecting a specific optimal muscle coordination pattern during gait. The proposed research will advance the computational environment at the Stanford Center for Biomedical Computation by providing basic surrogate modeling algorithms that are potentially applicable to other multiscale physics-based problems and also extend Center's efforts in neuromuscular biomechanics. [unreadable] [unreadable] [unreadable]